PIGS: Peridynamics-Informed Gaussian Splatting for Physics-Based Object Damage Modeling

Accurately modeling dynamic damage in 3D assets remains a core challenge for interactive digital environments, requiring a seamless integration of physically plausible fracture mechanics with high-fidelity novel-view synthesis. Existing physics-driven Gaussian Splatting approaches typically couple visual primitives with computational particles, inevitably compromising physical accuracy due to the highly non-uniform spatial distribution of Gaussian primitives. We introduce PIGS (Peridynamics-Informed Gaussian Splatting), a novel framework that fundamentally decouples the visual representation from the physical simulation. Our method leverages Peridynamics—a nonlocal continuum theory inherently adept at capturing spontaneous crack initiation and propagation—to govern the underlying material evolution, while independently optimizing the Gaussian primitives for rendering. To bridge these two spaces, we design a mapping scheme that translates time-varying physical states directly into updates of the Gaussian parameters, ensuring that the rendered imagery remains photorealistic and physically faithful throughout the entire damage sequence. This decoupled architecture not only preserves rigorous mechanical consistency but also enables rendering of complex damage processes. PIGS establishes a new paradigm for damage-aware digital twins, effectively unifying computational solid mechanics with differentiable 3D reconstruction.